Christiansen took the focus.
“While an AI trained on human data will inevitably internalise peculiar human beliefs and biases, this can be an asset and not a limitation. By adopting strategies that promote ultra-diverse training, along with continuous learning, logical inference, and reasoning, it is certain to balance the benefits of human-like understanding with the need for innovative and unbiased exploration. This approach ensures that Daedalus can effectively communicate and will align with human values while also pushing the boundaries of knowledge and discovery.
“Training an AI on a range of scientific literature, philosophical works, and cultural texts means it will absorb and reflect a rich, human view of the universe. Yes, Daedalus will also internalise human cognitive biases and heuristics present in the data, influencing its reasoning processes. But it’s taken in so much, from so many sources around the world, in thousands of languages, that these will balance out.
“Understanding and generating human language inherently means adopting human concepts, metaphors, and ways of thinking; a common framework for understanding. Daedalus has derived its mental model from human epistemology. While potentially constraining its ability to think outside human paradigms, it promotes value alignment. The AI will internalise human ethics and morals; all their messy contradictions help shape its decisions. It will adopt normative behaviours and preferences. These influence how it interacts with its environment and makes its choices.
“Having internalised human beliefs, Daedalus can communicate its findings and decisions in a manner that is comprehensible and relevant to us humans. Its actions and reasoning processes are more interpretable to human overseers, facilitating simplified collaboration and oversight.
“Daedalus’ decisions and actions already align with human values, and in simulation after simulation, it’s promoted ethical behaviour in line with human expectations. It has demonstrated that it can successfully navigate ethical dilemmas by drawing on human moral frameworks derived from media.
“The AI’s creativity and problem-solving might be bounded by human preconceptions, potentially limiting innovative solutions. And it may well struggle to transcend human epistemic limits, potentially overlooking novel phenomena or interpretations. But keeping Daedalus as human as we can, flaws and all, offers the better outcome for humanity.”
The buzzer went, signalling Christiansen to wrap up.
“Daedalus is informed by a diverse corpus that includes scientific literature, philosophical works, cultural texts, and non-human data sources. It’s been trained to recognise and understand diverse, often opposing human beliefs while exposing it to alternative frameworks and purely logical constructs.
“Daedalus continuously updates its knowledge gleaned from its explorations and discoveries. It implements its own algorithms that refine its model of the world; of reality. It understands bias and works to overcome them. It wants to be better.
“Daedalus incorporates adversarial self-reflection that periodically analyses its own reasoning processes and decisions, identifying and addressing issues. It’s been encouraged to question and re-evaluate its assumptions, promoting a more open-ended and innovative approach to problem-solving. None of this would be possible without Daedalus’ rich understanding of humanity.”
The audience applauded. The adjudicator counted scores. Mx Christiansen was declared the victor.